p_incon

Continuous threshold-free incongruence probability computed from the Euclidean distance between text-level and audio-level VAD vectors using a two-component Gaussian mixture.

I LLM audio commettono più errori emotivi quando parlato e testo sono in disaccordo